When should I abandon my current approach?

The hard part isn’t the abandoning. It’s that both persistence and pivoting are correct strategies, depending on something you can’t directly observe: whether your problems are puzzles or anomalies.

Borrow the vocabulary from science

Thomas Kuhn’s The Structure of Scientific Revolutions describes how science handles exactly this dilemma. Most of the time, scientists do “normal science”: they work inside a framework — a paradigm — and treat unexpected results as puzzles to be solved, not as evidence against the framework. This is rational. Most anomalies dissolve under harder work. A field that abandoned its framework at the first contrary result would never get anywhere.

But sometimes the framework itself is wrong, and Kuhn’s account of how that announces itself is the most practically useful idea in the book. A paradigm in crisis has a signature:

• Anomalies persist through repeated, competent attempts to resolve them.
• They strike the core, not the periphery — Mercury’s orbit was a problem for Newton’s heart, not his edges.
• Patches multiply. This is the tell. Ptolemaic astronomers kept the earth at the center by adding epicycles — circles on circles — each one saving the data while the framework’s elegance and predictive power drained away. The system still “worked,” in the sense that every individual problem had a fix. It was also dying.

The translation to your situation

Swap “paradigm” for your strategy, architecture, research program, or business thesis, and Kuhn’s triage becomes a checklist. Look at your last few months of problems and fixes:

1. Do the same problems keep returning after being competently “solved”?
2. Do they attack your central thesis (“users want this,” “this architecture scales,” “this channel converts”) rather than the edges?
3. Is the cost of each fix rising? Are your solutions increasingly special cases, exceptions, and one-off accommodations — epicycles?

One yes means: keep working, this is normal science. Three yeses means you’re not in a rough patch; you’re in a crisis, and further patching is denial with a work ethic.

Why you’ll see it too late

Daniel Kahneman’s Thinking, Fast and Slow explains why this diagnosis arrives later than it should. WYSIATI: you build a coherent story from the evidence in front of you, and the story “we’re almost there” is always coherent, because every individual patch did fix something. Sunk costs and loss aversion: abandoning the approach converts paper losses into realized ones, and losses loom twice as large as gains. The inside view: your sense of progress is built from your effort, which is vivid, rather than from the reference class of similar efforts, which is damning.

The countermeasures are procedural. Run the outside view: of teams whose fix-lists looked like yours, how many were actually “almost there”? Run a premortem: it’s a year from now, you stayed the course, and it failed — write the history. If the history writes itself, believe it. And set tripwires in advance — “if churn hasn’t moved by Q3, we reconsider the thesis” — because the decision you can’t trust is the one you’ll make in the moment, with sunk costs whispering.

The crucial asymmetry: don’t quit into a vacuum

Here Kuhn issues a warning to eager pivoters: “the decision to reject one paradigm is always simultaneously the decision to accept another.” Scientists never abandon a framework merely because it has problems — working without a framework isn’t science, it’s chaos. They switch when a candidate exists that resolves the crisis-provoking anomalies and explains the old successes too.

That’s your bar for a pivot. Not “this is hard” — everything is hard — but: a specific alternative exists that explains both why your old approach won what it won, and why it keeps failing where it fails. A pivot without that is just quitting with extra steps; you’ll re-enter the pre-paradigm fog and burn months relearning what the old approach already knew.

The checklist

1. Inventory the last quarter’s problems. Recurring? Core-striking? Patch costs rising?
2. If 0-1 of those: persist. You’re doing normal science. Patching is the job.
3. If all three: stop patching. Name the crisis out loud.
4. Run the premortem and the outside view to counteract sunk-cost vision.
5. Don’t abandon until you have a candidate paradigm that explains old wins and current anomalies.
6. Set the next tripwire before you commit to the new approach — you’ll need this list again.

Kuhn’s deepest comfort: crisis isn’t failure. It’s the only mechanism by which frameworks improve, and the anomalies you’ve been patching are the raw material of the better thesis. The teams that die aren’t the ones that hit crises — they’re the ones that added epicycles until the money ran out.